We investigated the real-time processing of two types of empty subject sentences. In a “subject-oriented sentence” such as “Taroo-ga Hanako-ni Tokyo iki-o hakuzyoosita (Taroo confessed to Hanako that he would go to Tokyo),” the person who will go to Tokyo is the subject Taroo . On the other hand, in an “object-oriented sentence” such as “Taroo-ga Hanako-ni Tokyo iki-o meireisita (Taroo ordered Hanako, (saying) she would go to Tokyo),” the object Hanako will go to Tokyo. Oda et al. (1997) found a “subject preference” phenomenon, since the main clause subject was preferred as the default antecedent for the empty subject. Furthermore, it was suggested that the human parser made top-down prediction at the beginning, and then changed to bottom-up processing by 900 msec after the end of the sentence. The stimuli were given after the end of each sentence. Consequently, it remained open whether the participants make any prediction while listening a sentence. By presenting the stimuli at six checkpoints prior to the end of the sentence, thus, the present study examined the transition of processing before the final verb has been shown. The overall results of the experiment revealed the following two findings: (i) “Yes” responses were faster than “No” responses in the subject-oriented sentences. Thus, it was assumed that the parser adopted a “subject preference” as a default strategy. (ii) The response times drastically increased at the checkpoint preceding the final verb. Therefore, it was suggested that the parser started to process a sentence before the final verb appeared.